Multivariate models of longitudinal data
We next considered service/medication patterns across all three waves in multivariate linear models that take into account site, gender and age (centered at 10), as well as diagnostic and risk variables. These models help to estimate the relative importance of these predictor sets on service, medication and psychostimulant use in each of the sites. In we present results from two versions of the model for each of these outcomes. The first considers only demographic and diagnostic variables, while the second adjusts for known child/family risk variables.
Multivariate Prediction of Service, Medication and Psychostimulant Use in Puerto Rican Children
After adjusting for demographic variables and considering the data from all time points, the site difference in overall service use remains statistically significant (b=−.63, se=0.11, p<.001). Children in the SB are 1.88 more likely to report service use than those in PR. Across both sites there was no evidence that service use varied over wave of study, but males were more likely to use service and age at baseline was related to service use in a nonlinear pattern. Children who were younger at baseline were less likely to use services but by age 10 there was little difference in the likelihood of service use with increasing age. Those with ADHD and ADHD-NOS were approximately three times more likely to use service than those with no ADHD (b=1.22, se=.18, p<.001, OR=3.4; b=1.08, se=.16, p<.001, OR=2.9), but the ADHD and ADHD-NOS groups did not differ significantly from each other. Those with another diagnosis besides ADHD also reported receiving more service, even after adjusting for comorbid ADHD (b=0.71, se=0.17, OR=2.0). also shows that children whose parents would consider medication use for their treatment were also more likely to receive services (b=0.48, se=.12, p<.001, OR=1.6).
The second multivariate model adjusted for the five risk variables in addition to the variables just reviewed. Four of these five variables were related to reported service use. Children in families with elevated Negative Family Influence scores were less likely to receive services (b=−0.21, se=.0.07, p<.01), while those exposed to elevated Environmental Risks (b=0.20, se=0.05, p<.0001), who themselves had elevated Negative Child Characteristics (b=0.77, se=0.07, p<.0001) and who had reduced Maternal Warmth (b=0.27, se=0.06, p<.0001) were all more likely to receive mental health services. These associations were all adjusted for ADHD diagnosis. However, it is interesting to note that the magnitudes of the associations of ADHD and ADHD-NOS with service use were dramatically reduced when adjusting for these risk processes. After adjustment, ADHD was no longer significant (b=0.34, se=.21, ns, OR=1.4) and ADHD-NOS was reduced by a third (b=0.68, se=.17, p<.001, OR=2.0). The adjustment also eliminated the association of other diagnoses (b=0.19, se=0.20, ns) and reduced the difference between the sites by half (b=−0.34, se=0.13, p<.01, OR=0.72).
Next we consider multivariate models of medication use. Model 1 in reveals substantial site (b=−1.12, se=0.22, p<.0001, OR=0.33) and gender differences (b=−1.0, se=0.24, p<.0001, OR=0.37) in medication use, with less use in PR and less use for females at both sites. There was no change over follow-up time in medication use, but there was a pattern with age at time of recruitment that resembled the pattern for overall service use. Children with ADHD (b=0.75, se=0.28, p<.01, OR=2.12) and ADHD-NOS (b=1.13, se=0.30, p<.001, OR=3.10) were more likely to receive medications, as were those with other diagnoses (b=0.65, se=0.30, p<.05, OR=1.92). Medication use was strongly related to caretakers’ reports of openness to pharmaceutical treatment, both stated unequivocally (“Yes”)(b=2.33, se=0.26, p<.001, OR=10.28) and ambivalently (“Maybe”)(b=0.74, se=0.34, p<.05, OR=2.10). When the five child/family risk variables were adjusted in Model 2, the associations of medication use to ADHD diagnoses (ADHD: b=0.19, se=0.34, ns; ADHD-NOS: b=0.18, se=.28, ns) and to other diagnoses (b=0.10, se=.32, ns) were again reduced to non-significance. In contrast, the effects for site, gender, and caretaker attitudes toward medication were not much affected by the adjustment. Among the five risk variables that were adjusted, three were related to medication use: Negative family influence (b=−0.32, se=0.12, p<.01), Negative Child Characteristics (b=0.61, se=0.13, p<.0001) and Lack of Maternal Warmth (b=0.50, se=0.11, p<.0001). Environmental risks, which had been significantly related to overall service use, had no association with medication use (b=0.09, se=0.10, ns).
The final two columns of show results for psychostimulant use (vs. no psychostimulant use). Like the results for medication use, psychostimulants are less used in PR (b=−0.96, se=0.28, p<.001, OR=0.38) and by girls (b=−1.0, se=0.32, p<.01, OR=0.37). ADHD (b=0.80, se=0.37, p<.05, OR=2.23) and AHDH-NOS (b=0.92, se=.34, p<.01, OR=2.51) are significantly related to psychostimulant use, but other diagnosis is not (b=0.60, se=0.37, ns, OR=1.82). Strong caretaker openness to pharmaceutical treatment is strongly related to psychostimulant use (b=2.53, se=0.36, p<.0001, OR=12.55) but ambivalent openness was not (b=0.52, se=0.50, ns, OR=1.68). In Model 2, Negative Family Influence (b=−0.36, se=0.16, p<.05), Negative Child Characteristics (b=0.50, se=0.16, p<.01) and Lack of Maternal Warmth (b=0.50, se=0.14, p<.001) were all related to psychostimulant use, but Ineffective Parent Structuring and Environmental Risks were not. When these risks were adjusted, the associations of ADHD and ADHD-NOS to psychostimulant use were reduced in magnitude and were no longer significant.
We reexamined these predictors of service and medication use in the sub-sample of children who received either the ADHD or ADHD-NOS diagnosis. These results are shown in . Overall, the patterns of results were similar to what we found in the total sample. Overall service was more likely to be provided to ADHD children in the SB than in PR, to males, to those who had comorbid disorders, and to those whose parents expressed openness to medication interventions. When individual risk variables were included, the diagnosis of a comorbid disorder no longer predicted service use. The same risk variables predict service use among ADHD children as in the total population: Exposure to environmental risks, negative child characteristics, and lack of maternal warmth made service use more likely, while negative family influences was inversely related to the likelihood of service use. Similar patterns were found for medication use. Important factors were site, gender, and caretaker attitude toward medication. Here again, comorbidity was not related to medication use once the sample is limited to those with ADHD.
Multivariate Prediction of Service, Medication and Psychostimulant Use in Puerto Rican Children with ADHD/ADHD-NOS at Baseline
The results for the ADHD subgroup were most different from the general sample for the prescription of medication in general and psychostimulants specifically. There is no evidence of a site difference in psychostimulant use, and the strongest predictor was caretaker’s attitude toward medication. Medication and psychostimulant use seem to be more a function of the caretaker’s attitude towards medication use than the psychiatric needs of the child. Children with ADHD whose caretakers were amenable to medication use (responding “yes” or “maybe” to their acceptance of using medications for their child’s behavior problems) were 11 times more likely to receive psychostimulants or other medications than those who were adverse to this type of intervention. Among caretakers of ADHD children receiving medication, an average (over three waves) of 90.5% in P.R. and 92.0% in the Bronx were amenable to medication use (“maybe” or “yes” responses) but the same was true of fewer caretakers (65.8% in PR and 48.0% in SB) of ADHD children not receiving medications. It is evident that the caretaker’s attitude towards medication use has a major bearing on whether an ADHD child receives medication.
Among the composite risk variables, only two were predictive. Those with ADHD were less likely to get psychostimulants if there was dysfunction in the family (Negative family influences), and they were more likely to get them if there was an impaired relationship with the caretaker (Low maternal acceptance/warmth). There was no indication that Negative child characteristics increased the likelihood that psychostimulants would be prescribed. The gender effect was diminished in this analysis, and was only significant when individual risk variables were adjusted.